Proximal robust factorization for piecewise planar reconstruction.

•A robust and practical factorization algorithm is proposed to recover 3D motion and dense planar scene.•A numerical scheme is developed to effectively integrate all the relevant constraints in a global objective function. The constraints include low rankness, scene smoothness and algebraic constraint associated with the shape matrix.•The algorithm can reliably remove outliers and yields a shape matrix that is meaningful and interpretable.